王洋-赋能研发创新-Databricks数据智能平台引领GenAI与智能Agent实践.pdf

1、赋能研发创新：Databricks数据智能平台引领GenAI与智能Agent实践王洋|Databricks中国架构师总监王洋(Will Wang)Databricks中国架构师总监超过15年的从业经验，涵盖大规模机器学习、湖仓平台及GenAI解决方案架构，致力于帮助数字原生企业与大型企业解决最复杂的数据与人工智能挑战。从机器学习工程师成长为解决方案架构师，再到如今的Databricks中国架构师团队负责人，Will兼具深厚的技术专长与敏锐的商业洞察，长期服务于制造、零售、Digital Native、金融服务与生命科学等多个行业客户，助力其实现数字化转型。在加入 Databricks 之前，W

2、ill曾在腾讯与 Cloudera 担任关键技术岗位，主导人工智能平台建设与大数据架构等核心项目。Will 热衷于用数据与AI将复杂问题转化为可扩展、可落地的解决方案，持续推动企业技术创新与业务增长。目 录CONTENTSI.Databricks 介绍II.Databricks Data Intelligence Platform 架构与能力总览III.构建 GenAI Agent 的端到端质量保障IV.MLflow 3 如何实现闭环质量保障V.行业落地实例Databricks 介绍PART 01The data and AI companyCreator of:10,000+global c

3、ustomers$2.4B+in annual revenue14B+in investmentInventor of thelakehouseand pioneer ofgenerative AILEADER2023 Cloud Database Management SystemsLEADER2024 Data Science&Machine LearningAnalytic Stream Processing DATAFORRESTER WAVE LEADER FOR DATA LAKEHOUSESAI2025 GARTNER DATA SCIENCE AND ML MQ+Databri

4、cks Data Intelligence Platform 架构与能力总览PART 02Disaster recoveryCost controlsEnterprise security100%serverlessLakehouseAI/BIBusiness intelligenceDatabricks SQLData warehousingWorkflows/DLTIngest,ETL,streamingMosaic AIArtificial intelligenceDatabricks Data Intelligence PlatformMosaic AI:The complete ag

5、ent platform Build agent systems that deliver accurate,domain-specific resultsGovern agentsAI guardrailsUsage trackingCredentialsRate limitsPrepare dataML featuresVector indexData ingestionBuild agentsGenAI modelsClassical ML modelsFunction and toolsDeploy agentsAgent servingMLOps/LLMOpsLineageEvalu

6、ate agentsLLM judgesPeer labelingTracingAgents that reason across every enterprise systemSupport for all existing and future AI modelsBuild trust with guardrails,evaluation,and monitoringLLMs maxing out on general intelligence tests20192020202120222023Open source vs.private models,5-Shot MMLU perfor